How AI search decides which businesses to name

When someone asks an AI tool "which solicitor in Oxford is good for probate?", the answer usually names 3 to 6 firms, says something specific about each and cites a handful of web pages. This guide explains, as far as the platforms have published it, how those names get chosen, and what we saw when we tested it.

No platform publishes its full method. Everything below comes either from official documentation, linked in the sources, or from our own measurements, which we label as such.

Step 1: the question gets broken down

Google says both AI Overviews and AI Mode "may use a query fan-out technique", issuing multiple related searches across subtopics and data sources to develop a response. A question such as "physio in Bristol for a running injury, open at weekends" can turn into separate searches about physios in Bristol, running injuries, weekend opening and prices.

The questions themselves are longer than traditional searches. When AI Mode launched in the UK on 28 July 2025, Google said early users were asking questions 2 to 3 times the length of traditional search queries.

For a business, that means each detail in a customer's question is a separate chance to match or miss. A page that says what you do, where, for whom, at what price and how soon covers more of those sub-questions than a page that just lists services.

Step 2: the tool looks for pages it can use

Each platform can only use pages it's allowed to reach and has already found.

• Google. A page must be indexed and eligible to be shown in Google Search with a snippet. Google says there are no additional technical requirements, and that robots.txt rules for Googlebot are how site owners control crawling for Search.
• ChatGPT search. OpenAI says OAI-SearchBot is used to surface websites in ChatGPT's search features. Sites that block it won't be shown in ChatGPT search answers, though they can still appear as navigational links. OpenAI also says ChatGPT search sometimes works with third-party search providers.
• Bing and Copilot. Microsoft's Bing Webmaster Tools now reports how often a site is cited in Copilot and AI-generated summaries in Bing, which confirms those answers draw on pages Bing has indexed.

The simplest failure is a site that blocks one of these crawlers by accident, often through a security setting at the hosting company or CDN. Google specifically tells site owners to check that their CDN or hosting set-up allows crawling.

Step 3: it reads what's on the page

Google's best-practice list for its AI features is short and practical: make important content available in text, support it with images and video, link to key pages from elsewhere on your site, make sure structured data matches the visible text, and keep Merchant Center and Business Profile information up to date.

Content that only exists in a PDF price list, an image of a menu or a booking widget is harder to use. When a price or a timetable is only in a format that's hard to read, AI either skips it or gets it wrong.

Step 4: it chooses who to name

This is the part platforms say least about. For local businesses, Google says its local results are mainly based on relevance, distance and prominence, and that complete, accurate Business Profile information, reviews and links all play a part. Complete profiles clearly feed AI answers too.

Our own research gives a picture of the outcome. On 13 October 2026 we asked Google AI Mode 12 questions a Londoner might ask about gyms, once each, logged out. Of 39 gym mentions, 28 (72%) were chains or groups and 11 (28%) were independents. 6 of the 12 answers named no independent gym at all, and 2 answers named a single gym. Read the research.

We can't see inside the model, so we can't say exactly why one business was chosen over another. What we can say is that the businesses named had published the specific facts the questions asked about, such as pools, saunas, opening hours and prices, and the answers repeated them.

Which sources answers rely on

Answers usually cite a mix of business websites, Google Business Profiles, directories and review sites. The website and the profile are the two a business controls, which is why our guide to Google Business Profile and our 20-point checklist start there.

Sources also go wrong. In our gym study, one of the source cards under an answer about King's Cross was a 24-hour gym in Reading, Berkshire, 40 miles away. Other sectors lean on different sources: B2B answers tend to cite comparison articles and review platforms, and trades answers may cite directories such as Checkatrade.

Why answers get facts wrong

Wrong facts usually trace back to a source that is unclear, out of date or in conflict with another source. When the website, the Google profile and a directory disagree on hours, prices or an address, the answer may pick any of them.

Why the same question gets different answers

AI answers aren't fixed. The same question can name different businesses on different days, for different users or in different locations. That's why a single screenshot, good or bad, proves little, and why we look across many answers on different days before reporting a number. Our guide to measuring AI search visibility covers the method.

What this means in practice

1. Check that Googlebot, Bingbot and OAI-SearchBot can reach your site.
2. Publish the specific facts customers ask about as text: services, prices, areas, people, hours and availability.
3. Give each main service its own page, so a detailed question has a detailed page to match.
4. Make your website, Google Business Profile and main listings say the same thing.
5. Ask every customer for an honest review in the same way, with no incentives.
6. Measure across many answers, and judge changes on counts.

If you'd like this done for you, see the services or ask for a free AI snapshot. The method is the same across every industry.